Radical-based HMM modeling for handwritten East Asian characters

Abstract

Exemplary methods, systems, and computer-readable media for developing, training and/or using models for online handwriting recognition of characters are described. An exemplary method for building a trainable radical-based HMM for use in character recognition includes defining radical nodes, where a radical node represents a structural element of an character, and defining connection nodes, where a connection node represents a spatial relationship between two or more radicals. Such a method may include determining a number of paths in the radical-based HMM using subsequence direction histogram vector (SDHV) clustering and determining a number of states in the radical-based HMM using curvature scale space-based (CSS) corner detection.

1. A method for character recognition, implemented at least in part by a computing device, the method comprising:
 receiving ink data for a character; 
 recognizing the character as associated with the received ink data using a radical-based Hidden Markov Model (HMM), the radical-based HMM comprising radical nodes, wherein a radical node represents a structural element of a character, and connection nodes, wherein a connection node represents a spatial relationship between two or more radicals; and 
 describing the radical-based HMM with a finite state machine that comprises at least one state selected from durative states or turning states. 
 
     
     
       2. The method of  claim 1  wherein the radical-based HMM comprises a multi-path topology wherein at least some paths of the multi-path topology traverse one or more radical nodes and one or more connection nodes. 
     
     
       3. The method of  claim 1  wherein the radical nodes represent radicals in a contextual radical set. 
     
     
       4. The method of  claim 3  wherein the contextual radical set accounts for shape variance of radicals with respect to characters. 
     
     
       5. The method of  claim 1  wherein the characters comprise East Asian characters. 
     
     
       6. A method for character recognition, implemented at least in part by a computing device, the method comprising:
 receiving ink data for a character; 
 recognizing the character as associated with the received ink data using a radical-based Hidden Markov Model (HMM), the radical-based HMM comprising radical nodes, wherein a radical node represents a structural element of the character, and connection nodes, wherein a connection node represents a spatial relationship between two or more radicals; and 
 describing the radical-based HMM with a finite state machine that comprises at least one state selected from durative states or turning states, wherein a durative state represents a stroke forming action in forming the character. 
 
     
     
       7. A method for character recognition, implemented at least in part by a computing device, the method comprising:
 receiving ink data for a character; 
 recognizing the character as associated with the received ink data using a radical-based Hidden Markov Model (HMM), the radical-based HMM comprising radical nodes, wherein a radical node represents a structural element of the character, and connection nodes, wherein a connection node represents a spatial relationship between two or more radicals; and 
 describing the radical-based HMM with a finite state machine that comprises at least one state selected from durative states or turning states, wherein a turning state represents a turning action in forming a character. 
 
     
     
       8. A method for training a radical-based Hidden Markov Model (HMM) for character recognition, implemented at least in part by a computing device, the method comprising:
 providing an initial radical-based HMM that comprises radical nodes and connection nodes; 
 splitting character ink data into radical data and connection data using the initial radical-based HMM; 
 training radical HMMs with the radical data and training connection HMMs with the connection data; 
 generating a trained radical-based HMM by concatenating the trained radical HMMs and the trained connection HMMs; and 
 determining a number of paths for the trained radical-based HMM, wherein the determining the number of paths for the trained radical-based HMM includes using a subsequence direction histogram vector (SDHV) clustering. 
 
     
     
       9. The method of  claim 8  further comprising iteratively training the radical-based HMM using the character ink data. 
     
     
       10. The method of  claim 8  further comprising splitting character ink data into radical data and connection data using the trained radical-based HMM. 
     
     
       11. The method of  claim 10  comprising generating a refined trained radical-based HMM using the radical data and the connection data split using the trained radical-based HMM. 
     
     
       12. The method of  claim 8  wherein the characters comprise East Asian characters. 
     
     
       13. A method for training a radical-based Hidden Markov Model (HMM) for character recognition, implemented at least in part by a computing device, the method comprising:
 providing an initial radical-based HMM that comprises radical nodes and connection nodes; 
 splitting character ink data into radical data and connection data using the initial radical-based HMM; 
 training radical HMMs with the radical data and training connection HMMs with the connection data; 
 generating a trained radical-based HMM by concatenating the trained radical HMMs and the trained connection HMMs; and 
 determining a number of states for the radical-based HMM, wherein the determining the number of states for the radical-based HMM comprises using a curvature scale space-based (CSS) corner detection. 
 
     
     
       14. A method for training a radical-based Hidden Markov Model (HMM) for character recognition, implemented at least in part by a computing device, the method comprising:
 providing an initial radical-based HMM that comprises radical nodes and connection nodes by:
 selecting a set of characters; 
 providing a set of radicals that can represent the characters; 
 providing types of connections that represent relationships between two or more radicals of the set of radicals; and 
 generating the initial radical-based HMM by constructing paths through nodes that represent radicals and nodes that represent types of connections using a path splitting algorithm that applies a convergence measure, wherein the convergence measure comprises self-rotation probabilities and leaving transition probabilities; 
 
 splitting character ink data into radical data and connection data using the initial radical-based HMM; 
 training radical HMMs with the radical data and training connection HMMs with the connection data; and 
 generating the trained radical-based HMM by concatenating the trained radical HMMs and the trained connection HMMs. 
 
     
     
       15. A method for training a radical-based Hidden Markov Model (HMM) for character recognition, implemented at least in part by a computing device, the method comprising:
 providing an initial radical-based HMM that comprises radical nodes and connection nodes by:
 selecting a set of characters; 
 providing a set of radicals that can represent the characters; 
 providing types of connections that represent relationships between two or more radicals of the set of radicals; and 
 generating the initial radical-based HMM by constructing paths through nodes that represent radicals and nodes that represent types of connections using a path splitting algorithm that applies a convergence measure; 
 
 splitting character ink data into radical data and connection data using the initial radical-based HMM; 
 training radical HMMs with the radical data and training connection HMMs with the connection data; and 
 generating the trained radical-based HMM by concatenating the trained radical HMMs and the trained connection HMMs.